[Gllug] Building Intranet Kiosks

[Nix]

On Thu, 12 Jul 2001, Kieran Barry uttered the following:
> On 12 Jul 2001, Nix wrote:
>> (btw *no* /sbin/halt bothers to fsck; do you mean `doesn't bother to
>> unmount'?)
>> 
> I don't know a great deal about NFS, so I guess he runs the risk of lots
> of stale connections?

(Assuming no NFS locking) the NFS server on the distant machine has no
knowledge of what NFS mounts the client has got; so there'd be no
staleness at all.

(I hope a diskless client isn't acting as an NFS server, just as a
client!)

>> >   sync ();
>> >   sleep (1);
>> >   sync ();
>> >   sleep (1);
>> >   sync ();
>> 
> You have to admit that a ~2.5 second shutdown is pretty attractive.

True --- and it is true that sync() isn't required to actually *finish*
disk writes, just schedule them. (In practice sane Unix systems try hard
to actually *do* the writes before the call returns, but this is
impossible with disk devices that do not report when data has actually
hit the disk.)

But no sleep() is long enough to be sure that a sync() is finished. A
umount() is the only way to be sure that everything is clean. (Although,
again, that isn't standardized, thanks to POSIX & friends airbrushing
over --- hell, completely ignoring --- mount(), umount() &c...)

> Don't know how long it takes to boot over TFTP and NFS tho.

Probably not very long, assuming an uncongested network --- and a
dedicated kiosknet would probably be quiet of all but kiosk-related
traffic. We're not talking Brunel University SlothNet(TM) speeds here.

-- 
`I'm not sure whether libtool is an existence proof that you _can_
 write a shell script that handles its arguments correctly, or a
 demonstration that you may try but you are doomed to failure.'
                                                       -- Zack Weinberg


-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug